Oblique-view video tracking and density-based counting: accurate counting of late-stage rapeseed seedlings for breeding assessment.

Accurate counting of late-stage rapeseed seedlings is critical for yield estimation and field management, while traditional manual counting is inefficient and labor-intensive, calling for an automated counting method. A novel video tracking and counting method (CropTriangulator) was proposed, which uses smartphone-captured videos to achieve row-based accurate counting based on oblique view and target density distribution. It integrates three core components: YOLOv11n was selected for its balanced detection accuracy and inference speed after model comparison; an adaptive DBSCAN (AdapDBSCAN) algorithm was designed to eliminate non-target seedlings by dynamically adjusting parameters to address perspective distortion; the SORT algorithm was adopted for tracking and counting, with permanent ID marking to ensure uniqueness when seedlings cross frame boundaries. Experiments on 20 test videos (10 for 45° oblique view, 10 for 90° vertical view) showed that CropTriangulator achieved an average counting accuracy of 97.13% at 45° (14% higher than 90°), with the R-squared of 45° row-based counts reaching 0.917. AdapDBSCAN reduced over-counting compared with fixed-parameter DBSCAN, and SORT had a much lower ID switch rate (8.47%) than DeepSORT (36.05%). The 45° oblique view is proven optimal for rapeseed seedling counting. The proposed CropTriangulator provides a low-cost and efficient solution for automated row-based counting in complex field environments, supporting precise yield estimation and scientific field management decisions. The video comparing the effects of the CropTriangulator method is available at: https://github.com/Possibility007/Comparison-of-counting-results.git.